Shuzheng Si

Haozhe Zhao, Zefan Cai, Shuzheng Si et al. · pku, stanford

Since the resurgence of deep learning, vision-language models (VLMs) enhanced by large language models (LLMs) have grown exponentially in popularity. However, while LLMs can utilize extensive background knowledge and task information with in-context learning, most VLMs still struggle with understanding complex multi-modal prompts with multiple images, making VLMs less effective in downstream vision-language tasks. In this paper, we address the limitation above by 1) introducing vision-language Model with Multi-Modal In-Context Learning(MMICL), a new approach to allow the VLM to deal with multi-modal inputs efficiently; 2) proposing a novel context scheme to augment the in-context learning ability of the VLM; 3) constructing the Multi-modal In-Context Learning (MIC) dataset, designed to enhance the VLM's ability to understand complex multi-modal prompts. Our experiments confirm that MMICL achieves new state-of-the-art zero-shot performance on a wide range of general vision-language tasks, especially for complex benchmarks, including MME and MMBench. Our analysis demonstrates that MMICL effectively tackles the challenge of complex multi-modal prompt understanding and emerges the impressive ICL ability. Furthermore, we observe that MMICL successfully alleviates language bias in VLMs, a common issue for VLMs that often leads to hallucination when faced with extensive textual context. Our code, dataset, dataset tool, and model are available at https://github.com/PKUnlp-icler/MIC

Xiangru Tang, Yuliang Liu, Zefan Cai et al. · pku

Despite Large Language Models (LLMs) like GPT-4 achieving impressive results in function-level code generation, they struggle with repository-scale code understanding (e.g., coming up with the right arguments for calling routines), requiring a deeper comprehension of complex file interactions. Also, recently, people have developed LLM agents that attempt to interact with repository code (e.g., compiling and evaluating its execution), prompting the need to evaluate their performance. These gaps have motivated our development of ML-Bench, a benchmark rooted in real-world programming applications that leverage existing code repositories to perform tasks. Addressing the need for LLMs to interpret long code contexts and translate instructions into precise, executable scripts, ML-Bench encompasses annotated 9,641 examples across 18 GitHub repositories, challenging LLMs to accommodate user-specified arguments and documentation intricacies effectively. To evaluate both LLMs and AI agents, two setups are employed: ML-LLM-Bench for assessing LLMs' text-to-code conversion within a predefined deployment environment, and ML-Agent-Bench for testing autonomous agents in an end-to-end task execution within a Linux sandbox environment. Our findings indicate that while GPT-4o leads with a Pass@5 rate surpassing 50%, there remains significant scope for improvement, highlighted by issues such as hallucinated outputs and difficulties with bash script generation. Notably, in the more demanding ML-Agent-Bench, GPT-4o achieves a 76.47% success rate, reflecting the efficacy of iterative action and feedback in complex task resolution. Our code, dataset, and models are available at https://github.com/gersteinlab/ML-bench.

Haozhe Zhao, Shuzheng Si, Zhenhailong Wang et al.

Scientific figures are among the most effective means of communicating complex research ideas, yet producing publication-quality illustrations remains one of the most labor-intensive parts of paper preparation. Existing automated systems each target a single figure type under text-only input, leaving the diversity of types and conditions researchers actually use unaddressed; their raster outputs further cannot be locally revised. Because scientific figures are structured compositions of discrete semantic components, the localized errors generators produce on such layouts demand not a stronger backbone but a harness. We instantiate this harness in two complementary systems: Crafter, a multi-agent harness for figure generation that generalizes across figure types and input conditions without architectural changes, and CraftEditor, which applies the same pattern to convert raster outputs into editable SVGs. Moreover, we introduce CraftBench, a benchmark spanning three figure types and four input conditions with human quality annotation. Experiments show that Crafter substantially outperforms both standalone generators and the agentic baseline on PaperBanana-Bench and CraftBench, with ablations confirming each component's independent contribution; CraftEditor faithfully converts outputs into editable SVGs that surpass all baselines. Our code and benchmark are available at https://github.com/HaozheZhao/Crafter.

Haozhe Zhao, Xiaojian Ma, Liang Chen et al. · pku

This paper presents UltraEdit, a large-scale (approximately 4 million editing samples), automatically generated dataset for instruction-based image editing. Our key idea is to address the drawbacks in existing image editing datasets like InstructPix2Pix and MagicBrush, and provide a systematic approach to producing massive and high-quality image editing samples. UltraEdit offers several distinct advantages: 1) It features a broader range of editing instructions by leveraging the creativity of large language models (LLMs) alongside in-context editing examples from human raters; 2) Its data sources are based on real images, including photographs and artworks, which provide greater diversity and reduced bias compared to datasets solely generated by text-to-image models; 3) It also supports region-based editing, enhanced by high-quality, automatically produced region annotations. Our experiments show that canonical diffusion-based editing baselines trained on UltraEdit set new records on MagicBrush and Emu-Edit benchmarks. Our analysis further confirms the crucial role of real image anchors and region-based editing data. The dataset, code, and models can be found in https://ultra-editing.github.io.

Shuzheng Si, Shuang Zeng, Baobao Chang · pku

Incomplete utterance rewriting has recently raised wide attention. However, previous works do not consider the semantic structural information between incomplete utterance and rewritten utterance or model the semantic structure implicitly and insufficiently. To address this problem, we propose a QUEry-Enhanced Network (QUEEN). Firstly, our proposed query template explicitly brings guided semantic structural knowledge between the incomplete utterance and the rewritten utterance making model perceive where to refer back to or recover omitted tokens. Then, we adopt a fast and effective edit operation scoring network to model the relation between two tokens. Benefiting from proposed query template and the well-designed edit operation scoring network, QUEEN achieves state-of-the-art performance on several public datasets.

Shuzheng Si, Shuang Zeng, Jiaxing Lin et al. · pku

Named Entity Recognition is the task to locate and classify the entities in the text. However, Unlabeled Entity Problem in NER datasets seriously hinders the improvement of NER performance. This paper proposes SCL-RAI to cope with this problem. Firstly, we decrease the distance of span representations with the same label while increasing it for different ones via span-based contrastive learning, which relieves the ambiguity among entities and improves the robustness of the model over unlabeled entities. Then we propose retrieval augmented inference to mitigate the decision boundary shifting problem. Our method significantly outperforms the previous SOTA method by 4.21% and 8.64% F1-score on two real-world datasets.

Shuzheng Si, Helan Hu, Haozhe Zhao et al. · pku

Distantly-Supervised Named Entity Recognition (DS-NER) is widely used in real-world scenarios. It can effectively alleviate the burden of annotation by matching entities in existing knowledge bases with snippets in the text but suffer from the label noise. Recent works attempt to adopt the teacher-student framework to gradually refine the training labels and improve the overall robustness. However, these teacher-student methods achieve limited performance because the poor calibration of the teacher network produces incorrectly pseudo-labeled samples, leading to error propagation. Therefore, we propose: (1) Uncertainty-Aware Teacher Learning that leverages the prediction uncertainty to reduce the number of incorrect pseudo labels in the self-training stage; (2) Student-Student Collaborative Learning that allows the transfer of reliable labels between two student networks instead of indiscriminately relying on all pseudo labels from its teacher, and further enables a full exploration of mislabeled samples rather than simply filtering unreliable pseudo-labeled samples. We evaluate our proposed method on five DS-NER datasets, demonstrating that our method is superior to the state-of-the-art DS-NER methods.

Kaikai An, Shuzheng Si, Helan Hu et al. · pku

Semantic Parsing aims to capture the meaning of a sentence and convert it into a logical, structured form. Previous studies show that semantic parsing enhances the performance of smaller models (e.g., BERT) on downstream tasks. However, it remains unclear whether the improvements extend similarly to LLMs. In this paper, our empirical findings reveal that, unlike smaller models, directly adding semantic parsing results into LLMs reduces their performance. To overcome this, we propose SENSE, a novel prompting approach that embeds semantic hints within the prompt. Experiments show that SENSE consistently improves LLMs' performance across various tasks, highlighting the potential of integrating semantic information to improve LLM capabilities.

Yucheng Cai, Wentao Ma, Yuchuan Wu et al.

Recent research has shown that multi-task pre-training greatly improves the model's robustness and transfer ability, which is crucial for building a high-quality dialog system. However, most previous works on multi-task pre-training rely heavily on human-defined input format or prompt, which is not optimal in quality and quantity. In this work, we propose to use Task-based Automatic Prompt generation (TAP) to automatically generate high-quality prompts. Using the high-quality prompts generated, we scale the corpus of the pre-trained conversation model to 122 datasets from 15 dialog-related tasks, resulting in Universal Pre-trained Conversation Model (UniPCM), a powerful foundation model for various conversational tasks and different dialog systems. Extensive experiments have shown that UniPCM is robust to input prompts and capable of various dialog-related tasks. Moreover, UniPCM has strong transfer ability and excels at low resource scenarios, achieving SOTA results on 9 different datasets ranging from task-oriented dialog to open-domain conversation. Furthermore, we are amazed to find that TAP can generate prompts on par with those collected with crowdsourcing. The code is released with the paper.

Shuzheng Si, Qingyi Wang, Haozhe Zhao et al.

Recognizing whether outputs from large language models (LLMs) contain faithfulness hallucination is crucial for real-world applications, e.g., retrieval-augmented generation and summarization. In this paper, we introduce FaithLens, a cost-efficient and effective faithfulness hallucination detection model that can jointly provide binary predictions and corresponding explanations to improve trustworthiness. To achieve this, we first synthesize training data with explanations via advanced LLMs and apply a well-defined data filtering strategy to ensure label correctness, explanation quality, and data diversity. Subsequently, we fine-tune the model on these well-curated training data as a cold start and further optimize it with rule-based reinforcement learning, using rewards for both prediction correctness and explanation quality. Results on 12 diverse tasks show that the 8B-parameter FaithLens outperforms advanced models such as GPT-4.1 and o3. Also, FaithLens can produce high-quality explanations, delivering a distinctive balance of trustworthiness, efficiency, and effectiveness.

Kaikai An, Li Sheng, Ganqu Cui et al. · pku, tsinghua

Instruction-following made modern large language models (LLMs) helpful assistants. However, the key to taming LLMs on complex instructions remains mysterious, for that there are huge gaps between models trained by open-source community and those trained by leading companies. To bridge the gap, we propose a simple and scalable approach UltraIF for building LLMs that can follow complex instructions with open-source data. UltraIF first decomposes real-world user prompts into simpler queries, constraints, and corresponding evaluation questions for the constraints. Then, we train an UltraComposer to compose constraint-associated prompts with evaluation questions. This prompt composer allows us to synthesize complicated instructions as well as filter responses with evaluation questions. In our experiment, for the first time, we successfully align LLaMA-3.1-8B-Base to catch up with its instruct version on 5 instruction-following benchmarks without any benchmark information, using only 8B model as response generator and evaluator. The aligned model also achieved competitive scores on other benchmarks. Moreover, we also show that UltraIF could further improve LLaMA-3.1-8B-Instruct through self-alignment, motivating broader use cases for the method. Our code is available at https://github.com/kkk-an/UltraIF.

Jinyang Li, Xiaolong Li, Ge Qu et al.

Resolution of complex SQL issues persists as a significant bottleneck in real-world database applications. Current Large Language Models (LLMs), while adept at text-to-SQL translation, have not been rigorously evaluated on the more challenging task of debugging SQL issues. To address this gap, we introduce BIRD-CRITIC, a new SQL issue debugging benchmark comprising 530 PostgreSQL tasks (BIRD-CRITIC-PG) and 570 multi-dialect tasks (BIRD-CRITIC-Multi), distilled from authentic user issues and replayed within new environments to facilitate rigorous evaluation. Baseline evaluations underscore the task's complexity, with the leading reasoning model O3-Mini achieving only 38.87% success rate on BIRD-CRITIC-PG and 33.33% on BIRD-CRITIC-Multi. Meanwhile, advancing open-source models for database tasks is crucial for empowering local development while safeguarding data privacy. Therefore, we present Six-Gym (Sql-fIX-Gym), a training environment for elevating open-source model capabilities for SQL issue debugging. This environment leverages SQL-Rewind strategy, which automatically generates executable issue-solution datasets by reverse-engineering issues from verified SQLs. However, popular trajectory-based fine-tuning methods do not explore substantial supervisory signals. We further propose f-Plan Boosting, which extracts high-level debugging plans from SQL solutions, enabling teacher LLMs to produce 73.7% more successful trajectories for training. We integrate these components into an open-source agent, Bird-Fixer. Based on Qwen-2.5-Coder-14B, Bird-Fixer achieves 38.11% success rate on BIRD-CRITIC-PG and 29.65% on BIRD-CRITIC-Multi, surpassing leading proprietary models such as Claude-3.7-Sonnet and GPT-4.1, marking a significant step toward democratizing sophisticated SQL-debugging capabilities. The leaderboard and source code are available: https://bird-critic.github.io/

Kangyang Luo, Shuzheng Si, Yuzhuo Bai et al.

In the era of large language models (LLMs), supervised neural methods remain the state-of-the-art (SOTA) for Coreference Resolution. Yet, their full potential is underexplored, particularly in incremental clustering, which faces the critical challenge of balancing efficiency with performance for long texts. To address the limitation, we propose \textbf{MEIC-DT}, a novel dual-threshold, memory-efficient incremental clustering approach based on a lightweight Transformer. MEIC-DT features a dual-threshold constraint mechanism designed to precisely control the Transformer's input scale within a predefined memory budget. This mechanism incorporates a Statistics-Aware Eviction Strategy (\textbf{SAES}), which utilizes distinct statistical profiles from the training and inference phases for intelligent cache management. Furthermore, we introduce an Internal Regularization Policy (\textbf{IRP}) that strategically condenses clusters by selecting the most representative mentions, thereby preserving semantic integrity. Extensive experiments on common benchmarks demonstrate that MEIC-DT achieves highly competitive coreference performance under stringent memory constraints.

Haozhe Zhao, Zefan Cai, Shuzheng Si et al.

Recent text-to-image models produce high-quality results but still struggle with precise visual control, balancing multimodal inputs, and requiring extensive training for complex multimodal image generation. To address these limitations, we propose MENTOR, a novel autoregressive (AR) framework for efficient Multimodal-conditioned Tuning for Autoregressive multimodal image generation. MENTOR combines an AR image generator with a two-stage training paradigm, enabling fine-grained, token-level alignment between multimodal inputs and image outputs without relying on auxiliary adapters or cross-attention modules. The two-stage training consists of: (1) a multimodal alignment stage that establishes robust pixel- and semantic-level alignment, followed by (2) a multimodal instruction tuning stage that balances the integration of multimodal inputs and enhances generation controllability. Despite modest model size, suboptimal base components, and limited training resources, MENTOR achieves strong performance on the DreamBench++ benchmark, outperforming competitive baselines in concept preservation and prompt following. Additionally, our method delivers superior image reconstruction fidelity, broad task adaptability, and improved training efficiency compared to diffusion-based methods. Dataset, code, and models are available at: https://github.com/HaozheZhao/MENTOR

Haozhe Zhao, Zefan Cai, Shuzheng Si et al. · pku

Large-scale multilingual Pretrained Language Models (mPLMs) yield impressive performance on cross-language tasks, yet significant performance disparities exist across different languages within the same mPLM. Previous studies endeavored to narrow these disparities by supervise fine-tuning the mPLMs with multilingual data. However, obtaining labeled multilingual data is time-consuming, and fine-tuning mPLM with limited labeled multilingual data merely encapsulates the knowledge specific to the labeled data. Therefore, we introduce ALSACE to leverage the learned knowledge from the well-performing languages to guide under-performing ones within the same mPLM, eliminating the need for additional labeled multilingual data. Experiments show that ALSACE effectively mitigates language-level performance disparity across various mPLMs while showing the competitive performance on different multilingual NLU tasks, ranging from full resource to limited resource settings. The code for our approach is available at https://github.com/pkunlp-icler/ALSACE.

Shuzheng Si, Haozhe Zhao, Yu Lei et al.

Many real-world tasks require language models (LMs) to reason over complex contexts that exceed their parametric knowledge. This calls for context learning, where LMs directly learn relevant knowledge from the given context. An intuitive solution is inference-time skill augmentation: extracting the rules and procedures from context into natural-language skills. However, constructing such skills for context learning scenarios faces two challenges: the prohibitive cost of manual skill annotation for long, technically dense contexts, and the lack of external feedback for automated skill construction, since there is no automatic signal to tell whether a proposed skill is helpful. In this paper, we propose Ctx2Skill, a self-evolving framework that autonomously discovers, refines, and selects context-specific skills without human supervision or external feedback. At its core, a multi-agent self-play loop has a Challenger that generates probing tasks and rubrics, a Reasoner that attempts to solve them guided by an evolving skill set, and a neutral Judge that provides binary feedback. Crucially, both the Challenger and the Reasoner evolve through accumulated skills: dedicated Proposer and Generator agents analyze failure cases and synthesize them into targeted skill updates for both sides, enabling automated skill discovery and refinement. To prevent adversarial collapse caused by increasingly extreme task generation and over-specialized skill accumulation, we further introduce a Cross-time Replay mechanism that identifies the skill set achieving the best balance across representative cases for the Reasoner side, ensuring robust and generalizable skill evolution. The resulting skills can be plugged into any language model to obtain better context learning capability. Evaluated on four context learning tasks from CL-bench, Ctx2Skill consistently improves solving rates across backbone models.

Yunshui Li, Binyuan Hui, Xiaobo Xia et al. · tsinghua

Contemporary practices in instruction tuning often hinge on enlarging data scaling without a clear strategy for ensuring data quality, inadvertently introducing noise that may compromise model performance. To address this challenge, we introduce \textsc{Nuggets}, a novel and efficient methodology that leverages one-shot learning to discern and select high-quality instruction data from extensive datasets. \textsc{Nuggets} assesses the potential of individual instruction examples to act as effective one-shot learning instances, thereby identifying those that can significantly improve performance across diverse tasks. \textsc{Nuggets} utilizes a scoring system based on the impact of candidate examples on the perplexity of a diverse anchor set, facilitating the selection of the most advantageous data for instruction tuning. Through comprehensive evaluations on two benchmarks, including MT-Bench and Alpaca-Eval, we show that instruction tuning with the top 1\% of examples curated by \textsc{Nuggets} substantially outperforms conventional methods employing the entire dataset.

Haozhe Zhao, Shuzheng Si, Liang Chen et al.

Large vision-language models (LVLMs) have achieved impressive results in various vision-language tasks. However, despite showing promising performance, LVLMs suffer from hallucinations caused by language bias, leading to diminished focus on images and ineffective visual comprehension. We identify two primary reasons for this bias: 1. Different scales of training data between the pretraining stage of LLM and multimodal alignment stage. 2. The learned inference bias due to short-term dependency of text data. Therefore, we propose LACING, a systemic framework designed to address the language bias of LVLMs with muLtimodal duAl-attention meChanIsm (MDA) aNd soft-image Guidance (IFG). Specifically, MDA introduces a parallel dual-attention mechanism that enhances the integration of visual inputs across the model. IFG introduces a learnable soft visual prompt during training and inference to replace visual inputs, designed to compel LVLMs to prioritize text inputs. Then, IFG further proposes a novel decoding strategy using the soft visual prompt to mitigate the model's over-reliance on adjacent text inputs. Comprehensive experiments demonstrate that our method effectively debiases LVLMs from their language bias, enhancing visual comprehension and reducing hallucinations without requiring additional training resources or data. The code and model are available at [lacing-lvlm.github.io](https://lacing-lvlm.github.io).

Shuzheng Si, Haozhe Zhao, Gang Chen et al. · pku, tsinghua

Training LLMs on data containing unfamiliar knowledge during the instruction tuning stage can encourage hallucinations. To address this challenge, we introduce NOVA, a novel framework designed to identify high-quality data that aligns well with the LLM's learned knowledge to reduce hallucinations. NOVA includes Internal Consistency Probing (ICP) and Semantic Equivalence Identification (SEI) to measure how familiar the LLM is with instruction data. Specifically, ICP evaluates the LLM's understanding of the given instruction by calculating the tailored consistency among multiple self-generated responses. SEI further assesses the familiarity of the LLM with the target response by comparing it to the generated responses, using the proposed semantic clustering and well-designed voting strategy. Finally, to ensure the quality of selected samples, we introduce an expert-aligned reward model, considering characteristics beyond just familiarity. By considering data quality and avoiding unfamiliar data, we can utilize the selected data to effectively align LLMs to follow instructions and hallucinate less.

Shuzheng Si, Haozhe Zhao, Gang Chen et al. · pku, tsinghua

Aligning large language models to handle instructions with extremely long contexts has yet to be fully investigated. Previous studies have attempted to scale up the available data volume by synthesizing long instruction-following samples, as constructing such a dataset tends to be challenging for annotators. However, a lack of a well-defined strategy for ensuring data quality may introduce low-quality samples and restrict the model's performance. Thus, we propose GATEAU, a novel framework to address the unique challenge of long context alignment by identifying the influential samples enriched with long-range dependency relations. Specifically, GATEAU measures the long-range dependencies from two essential aspects: the difficulty of generating target responses due to the long-range dependencies, and the difficulty of understanding long inputs due to such dependencies. Comprehensive experiments indicate that GATEAU effectively identifies influential samples, and the model trained on these selected samples exhibits better instruction-following and long-context understanding capabilities.

Shuzheng Si, Haozhe Zhao, Cheng Gao et al. · tsinghua

Teaching large language models (LLMs) to be faithful in the provided context is crucial for building reliable information-seeking systems. Therefore, we propose a systematic framework, CANOE, to reduce faithfulness hallucinations of LLMs across different downstream tasks without human annotations. Specifically, we first synthesize short-form question-answering (QA) data with four diverse tasks to construct high-quality and easily verifiable training data without human annotation. Also, we propose Dual-GRPO, a rule-based reinforcement learning method that includes three tailored rule-based rewards derived from synthesized short-form QA data, while simultaneously optimizing both short-form and long-form response generation. Notably, Dual-GRPO eliminates the need to manually label preference data to train reward models and avoids over-optimizing short-form generation when relying only on the synthesized short-form QA data. Experimental results show that CANOE greatly improves the faithfulness of LLMs across 11 different tasks, even outperforming the most advanced LLMs, e.g., GPT-4o and OpenAI o1.

Kangyang Luo, Yuzhuo Bai, Cheng Gao et al.

Knowledge Graph Completion (KGC), which aims to infer missing or incomplete facts, is a crucial task for KGs. However, integrating the vital structural information of KGs into Large Language Models (LLMs) and outputting predictions deterministically remains challenging. To address this, we propose a new method called GLTW, which encodes the structural information of KGs and merges it with LLMs to enhance KGC performance. Specifically, we introduce an improved Graph Transformer (iGT) that effectively encodes subgraphs with both local and global structural information and inherits the characteristics of language model, bypassing training from scratch. Also, we develop a subgraph-based multi-classification training objective, using all entities within KG as classification objects, to boost learning efficiency.Importantly, we combine iGT with an LLM that takes KG language prompts as input.Our extensive experiments on various KG datasets show that GLTW achieves significant performance gains compared to SOTA baselines.

Cheng Gao, Cheng Huang, Kangyang Luo et al.

Enabling large language models (LLMs) to appropriately abstain from answering questions beyond their knowledge is crucial for mitigating hallucinations. While existing reinforcement learning methods foster autonomous abstention, they often compromise answer accuracy because their static reward mechanisms, agnostic to models' knowledge boundaries, drive models toward excessive caution. In this work, we propose KARL, a novel framework that continuously aligns an LLM's abstention behavior with its evolving knowledge boundary. KARL introduces two core innovations: a Knowledge-Boundary-Aware Reward that performs online knowledge boundary estimation using within-group response statistics, dynamically rewarding correct answers or guided abstention; and a Two-Stage RL Training Strategy that first explores the knowledge boundary and bypasses the "abstention trap", and subsequently converts incorrect answers beyond the knowledge boundary into abstentions without sacrificing accuracy. Extensive experiments on multiple benchmarks demonstrate that KARL achieves a superior accuracy-hallucination trade-off, effectively suppressing hallucinations while maintaining high accuracy across both in-distribution and out-of-distribution scenarios.

Nan Huo, Xiaohan Xu, Jinyang Li et al.

Large language models (LLMs) have demonstrated remarkable performance on single-turn text-to-SQL tasks, but real-world database applications predominantly require multi-turn interactions to handle ambiguous queries, execution errors, and evolving user requirements. Existing multi-turn benchmarks fall short by treating conversation histories as static context or limiting evaluation to read-only operations, failing to reflect production-grade database assistant challenges. We introduce BIRD-INTERACT, a benchmark that restores this realism through: (1) a comprehensive interaction environment coupling each database with a hierarchical knowledge base, metadata files, and a function-driven user simulator, enabling models to solicit clarifications, retrieve knowledge, and recover from errors without human supervision; (2) two evaluation settings consisting of a pre-defined conversational protocol (c-Interact) and an open-ended agentic setting (a-Interact) where models autonomously decide when to query the user simulator or explore the environment; (3) a challenging task suite covering the full CRUD spectrum for business-intelligence and operational use cases, guarded by executable test cases. Each task features ambiguous and follow-up sub-tasks requiring dynamic interaction. The suite comprises BIRD-INTERACT-FULL (600 tasks, up to 11,796 interactions) for comprehensive performance assessment, and BIRD-INTERACT-LITE (300 tasks with simplified databases) for detailed behavioral analysis and rapid method development. Our empirical results highlight BIRD-INTERACT's difficulty: GPT-5 completes only 8.67% of tasks in c-Interact and 17.00% in a-Interact. Analysis via memory grafting and Interaction Test-time Scaling validates the importance of effective interaction for complex, dynamic text-to-SQL tasks.

Yiqing Zhou, Yu Lei, Shuzheng Si et al.

Managing extensive context remains a critical bottleneck for Large Language Models (LLMs), particularly in applications like long-document question answering and autonomous agents where lengthy inputs incur high computational costs and introduce noise. Existing compression techniques often disrupt local coherence through discrete token removal or rely on implicit latent encoding that suffers from positional bias and incompatibility with closed-source APIs. To address these limitations, we introduce the EDU-based Context Compressor, a novel explicit compression framework designed to preserve both global structure and fine-grained details. Our approach reformulates context compression as a structure-then-select process. First, our LingoEDU transforms linear text into a structural relation tree of Elementary Discourse Units (EDUs) which are anchored strictly to source indices to eliminate hallucination. Second, a lightweight ranking module selects query-relevant sub-trees for linearization. To rigorously evaluate structural understanding, we release StructBench, a manually annotated dataset of 248 diverse documents. Empirical results demonstrate that our method achieves state-of-the-art structural prediction accuracy and significantly outperforms frontier LLMs while reducing costs. Furthermore, our structure-aware compression substantially enhances performance across downstream tasks ranging from long-context tasks to complex Deep Search scenarios.

Yu Lei, Shuzheng Si, Wei Wang et al.

Large language models are evolving from single-turn responders into tool-using agents capable of sustained reasoning and decision-making for deep research. Prevailing systems adopt a linear pipeline of plan to search to write to a report, which suffers from error accumulation and context rot due to the lack of explicit control over both model behavior and context. We introduce RhinoInsight, a deep research framework that adds two control mechanisms to enhance robustness, traceability, and overall quality without parameter updates. First, a Verifiable Checklist module transforms user requirements into traceable and verifiable sub-goals, incorporates human or LLM critics for refinement, and compiles a hierarchical outline to anchor subsequent actions and prevent non-executable planning. Second, an Evidence Audit module structures search content, iteratively updates the outline, and prunes noisy context, while a critic ranks and binds high-quality evidence to drafted content to ensure verifiability and reduce hallucinations. Our experiments demonstrate that RhinoInsight achieves state-of-the-art performance on deep research tasks while remaining competitive on deep search tasks.

Kangyang Luo, Yuzhuo Bai, Shuzheng Si et al. · tsinghua

Coreference Resolution (CR) is a critical task in Natural Language Processing (NLP). Current research faces a key dilemma: whether to further explore the potential of supervised neural methods based on small language models, whose detect-then-cluster pipeline still delivers top performance, or embrace the powerful capabilities of Large Language Models (LLMs). However, effectively combining their strengths remains underexplored. To this end, we propose \textbf{ImCoref-CeS}, a novel framework that integrates an enhanced supervised model with LLM-based reasoning. First, we present an improved CR method (\textbf{ImCoref}) to push the performance boundaries of the supervised neural method by introducing a lightweight bridging module to enhance long-text encoding capability, devising a biaffine scorer to comprehensively capture positional information, and invoking a hybrid mention regularization to improve training efficiency. Importantly, we employ an LLM acting as a multi-role Checker-Splitter agent to validate candidate mentions (filtering out invalid ones) and coreference results (splitting erroneous clusters) predicted by ImCoref. Extensive experiments demonstrate the effectiveness of ImCoref-CeS, which achieves superior performance compared to existing state-of-the-art (SOTA) methods.

Shuzheng Si, Haozhe Zhao, Kangyang Luo et al. · tsinghua

Agents based on large language models (LLMs) struggle with brainless trial-and-error and generating hallucinatory actions due to a lack of global planning in long-horizon tasks. In this paper, we introduce a plan-and-execute framework and propose EAGLET, an efficient and effective planner training method to enhance the executor agent's planning abilities without human effort. Specifically, we train a plug-and-play global planner through a two-step process: we first synthesize high-quality plans from an advanced LLM using our proposed homologous consensus filtering strategy, and apply fine-tuning as a cold start. Moreover, we further improve the planner with a rule-based reinforcement learning stage using a novel executor capability gain reward, ensuring it can handle task instructions of varying difficulty. Experiments on three long-horizon agent tasks show that executor agents equipped with our planner outperform existing methods, achieving new state-of-the-art performance. Meanwhile, EAGLET reduces training costs by 8x compared to RL-based baselines, and it does not require manual effort or extra training data, offering an efficient and effective solution.

Kaikai An, Fangkai Yang, Liqun Li et al.

Recent advances in retrieval-augmented generation (RAG) have substantially improved question-answering systems, particularly for factoid '5Ws' questions. However, significant challenges remain when addressing '1H' questions, specifically how-to questions, which are integral for decision-making and require dynamic, step-by-step responses. The key limitation lies in the prevalent data organization paradigm, chunk, which commonly divides documents into fixed-size segments, and disrupts the logical coherence and connections within the context. To address this, we propose Thread, a novel data organization paradigm enabling systems to handle how-to questions more effectively. Specifically, we introduce a new knowledge granularity, 'logic unit' (LU), where large language models transform documents into more structured and loosely interconnected LUs. Extensive experiments across both open-domain and industrial settings show that Thread outperforms existing paradigms significantly, improving the success rate of handling how-to questions by 21% to 33%. Additionally, Thread demonstrates high adaptability across diverse document formats, reducing retrieval information by up to 75% compared to chunk, and also shows better generalizability to '5Ws' questions, such as multi-hop questions, outperforming other paradigms.

Shuzheng Si, Wentao Ma, Haoyu Gao et al.

Task-oriented dialogue (TOD) models have made significant progress in recent years. However, previous studies primarily focus on datasets written by annotators, which has resulted in a gap between academic research and real-world spoken conversation scenarios. While several small-scale spoken TOD datasets are proposed to address robustness issues such as ASR errors, they ignore the unique challenges in spoken conversation. To tackle the limitations, we introduce SpokenWOZ, a large-scale speech-text dataset for spoken TOD, containing 8 domains, 203k turns, 5.7k dialogues and 249 hours of audios from human-to-human spoken conversations. SpokenWOZ further incorporates common spoken characteristics such as word-by-word processing and reasoning in spoken language. Based on these characteristics, we present cross-turn slot and reasoning slot detection as new challenges. We conduct experiments on various baselines, including text-modal models, newly proposed dual-modal models, and LLMs, e.g., ChatGPT. The results show that the current models still have substantial room for improvement in spoken conversation, where the most advanced dialogue state tracker only achieves 25.65% in joint goal accuracy and the SOTA end-to-end model only correctly completes the user request in 52.1% of dialogues. The dataset, code, and leaderboard are available: https://spokenwoz.github.io/.

Shuzheng Si, Zefan Cai, Shuang Zeng et al.

Distantly-Supervised Named Entity Recognition effectively alleviates the burden of time-consuming and expensive annotation in the supervised setting. But the context-free matching process and the limited coverage of knowledge bases introduce inaccurate and incomplete annotation noise respectively. Previous studies either considered only incomplete annotation noise or indiscriminately handle two types of noise with the same strategy. In this paper, we argue that the different causes of two types of noise bring up the requirement of different strategies in model architecture. Therefore, we propose the SANTA to handle these two types of noise separately with (1) Memory-smoothed Focal Loss and Entity-aware KNN to relieve the entity ambiguity problem caused by inaccurate annotation, and (2) Boundary Mixup to alleviate decision boundary shifting problem caused by incomplete annotation and a noise-tolerant loss to improve the robustness. Benefiting from our separate tailored strategies, we confirm in the experiment that the two types of noise are well mitigated. SANTA also achieves a new state-of-the-art on five public datasets.